Yes, I've also seen the same problem with EA testing using strategy tester...for the same set-up, each run provides different outcome (sometimes no trades at all!).

This problem increases as you try to test your strategy on older data and also on exotic crosses (non-major pairs). I have posted a question in MetaQuotes forum, but so far no reply.

My current solution is to run the test multiple times (takes more time) and then eye-ball the average statistics to get an idea about how effective your EA is.